Daily Image
13-08-2010
Ghosts of Selfcal Past
| Submitter: | Oleg Smirnov |
| Description: | John Wheeler once famously said, "If you are not completely confused by quantum mechanics, you do not understand it." We don't know if he understood self-calibration, but we can only assume that if he did, he would have found it pretty confusing too! In 2004, Ger de Bruyn obtained a 92cm WSRT observation of the field J1819+3845. This turned up a string of "ghost sources" connecting the brightest source in the field with Cyg A, which was 20 degrees away, coming in through a far sidelobe. The ghosts were remarkable in that they were composed of perfect rings, and their position remained constant with frequency. Despite trying, nobody could come up with an adequate explanation for the ghosts -- it wasn't even clear if they were errors in the data, or were a calibration artifact. Eventually the mystery was filed away as a one-off error and mostly forgotten. Recently, we started seeing ghosts in LOFAR maps. Despite looking very different, they did bring back memories. Then, serendipity struck. A wonky telescope during a completely unrelated 21cm WSRT observation ( http://www.astron.nl/dailyimage/main.php?date=20100805 ) produced a string of ghosts that looked very similar to the 2004 ones. Even better, they went away once differential gains were applied and selfcal was redone. This proved to be the vital clue. It turns out that ghosts are caused, rather simply, by "selfcal contamination" -- i.e. flux missing from the sky model during calibration. Of course, contamination occurs (almost by definition) during any selfcal, but normally it manifests itself as faint artifacts around dominant sources, plus noise-like crud. It turns out that if the contaminating flux is localized, and is far away from the phase centre (as was the case in 2004, where an imperfect model of CygA was responsible for contamination), and the dynamic range is sufficiently high, the crud turns into coherent structures -- namely, ghosts! We do not yet understand all the mathematics involved (like quantum mechanics, the more we understand, the more fascinating and confusing they get), but a series of simulations makes this process perfectly clear. The animation above shows one result of our simulations. Here, we have simulated a 1 Jy central point source, and three 1 mJy "contaminators". We do selfcal using a model containing only the central source, which produces slightly incorrect solutions due to the unmodelled contaminators. We then subtract all sources from the data. The resulting image shows "distilled" artifacts: flux from the central source that has been scattered through the map by the application of contaminated selfcal solutions. Successive frames of the animation correspond to increasing distance of the contaminators from the phase centre. As they get further away, perfect counterparts to the 2004 ghosts start forming up! A more detailed report on these matters will be presented at the CALIM2010 meeting at ASTRON, on Monday, August 23rd. |
| Copyright: | (c) |
| Tweet |  |